Timeline for Partially Full Factorial Logistic Regression - Bias?
Current License: CC BY-SA 3.0
7 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Apr 13, 2017 at 12:44 | history | edited | CommunityBot |
replaced http://stats.stackexchange.com/ with https://stats.stackexchange.com/
|
|
| Feb 24, 2017 at 21:19 | history | bounty ended | B_Miner | ||
| Feb 24, 2017 at 17:40 | history | edited | eric_kernfeld | CC BY-SA 3.0 |
caveat RE: block matrices
|
| Feb 24, 2017 at 17:37 | comment | added | eric_kernfeld |
Yes. For any model with enough interactions to fit every combination of factor levels separately, the "main effects" in the default glm parameterization will be fitted to the first level of each factor. Think of a simple scenario: $\begin{bmatrix}1\\10\end{bmatrix} = \begin{bmatrix}1 , 0\\1,1\end{bmatrix}* \begin{bmatrix}1\\9\end{bmatrix}$, so the "main effect" is fitted to the first observation while the interaction makes up the difference to the second.
|
|
| Feb 24, 2017 at 1:23 | comment | added | B_Miner | "A smaller model on a subset yields the same results. Why?" is the main part of my question I think. I am still not sure why the main effect of each variable (i.e. for population A) are unaffected by the other rows. Is this the case with any linear model with an interaction term, that essentially it does a subset regression (for lack of a better term)? | |
| Feb 23, 2017 at 3:11 | history | edited | eric_kernfeld | CC BY-SA 3.0 |
Population A instead of population 0
|
| Feb 23, 2017 at 3:05 | history | answered | eric_kernfeld | CC BY-SA 3.0 |